This is how it was up to v2.1.2 included (changed in v2.1.4 onward).
I added a benchmark that has 3 subscribers running and increase
the number of publishers: 1, 2, 5 and 10. This is the comparison
between the pre-PR and post-PR:
```
benchcmp old.txt new.txt
benchmark old ns/op new ns/op delta
Benchmark___BumpPubCount_1x3-16 396 385 -2.78%
Benchmark___BumpPubCount_2x3-16 495 406 -17.98%
Benchmark___BumpPubCount_5x3-16 542 395 -27.12%
Benchmark__BumpPubCount_10x3-16 549 515 -6.19%
benchmark old MB/s new MB/s speedup
Benchmark___BumpPubCount_1x3-16 717.27 737.54 1.03x
Benchmark___BumpPubCount_2x3-16 574.31 699.02 1.22x
Benchmark___BumpPubCount_5x3-16 524.35 718.80 1.37x
Benchmark__BumpPubCount_10x3-16 517.26 551.53 1.07x
```
It is inline with what the user reported, seeing a 20% drop in performance
when going from 1 publisher to 2. But, as we can see, the difference
between go channel and cond variable reduces with the increased number
of publishers after a certain number.
I am not sure of the performance impact on other situations, so this
PR is more of a proposal than a fix.
Resolves#1786
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
API made more consistent. Noun followed by verb.
Name arguments in request subejcts are always at the end now.
Remove enabled call, just use account info.
Getting a message directly from a stream is treated like an admin API and requires JSON request.
Deleting a message directly as well.
StreamList and ConsumerList now include details and support paging.
Streams and Consumers now contain a created field in their info.
Signed-off-by: Derek Collison <derek@nats.io>
This is a breaking change and will not be able to restore consumer's from a filestore when upgraded.
We are getting close to settling on the API an once that happens we will not be introducnig any
breaking changes.
Signed-off-by: Derek Collison <derek@nats.io>
This is the first checkin for JetStream. Has some rudimentary basics working.
TODO
1. Push vs pull mode for observables. (work queues)
2. Disk/File store, memory only for now.
3. clustering code - design shaping up well.
4. Finalize account import semantics.
5. Lots of other little things.
Signed-off-by: Derek Collison <derek@nats.io>
- New prefix that includes origin server for the request
- Mapping done if request is service import or requestor has
recent subscription
- Subscription considered recent if less than 250ms
- Destination server strip GW prefix before giving to client
and restore when getting a reply on that subject
- Mapping removed aftert 250ms
- Server rejects client publish on "$GNR." (the new prefix)
- Cluster and server hash are now 8 chars long and from base 62
alphabets
- Mapped replies need to be sent to leafnode servers due to race
(cluster B sends RS+ on GW inbound then RMSG on outbound, the
RS+ may be processed later and cluster A may have given message
to LN before RS+ on reply subject. So LN needs to accept the
mapped reply but will strip to give to client and reassemble
before sending it back)
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
Allow disabling of short first ping timer for clients.
Adjust names so that full test suite results are aligned.
Removed the account lookup, we use sync.Map but also a no-lock cache.
Signed-off-by: Derek Collison <derek@nats.io>
With Go 1.12 (strangely was not able to reproduce with Go 1.11)
the test TestRouteNoCrashOnAddingSubToRoute() would frequently
locks up and consume all avail CPUs on the machine. Running
this test with GOMAXPROCS=2 you would see server.test CPU usage
pegged at 200% (assuming you have at least 2 CPUs).
The reason was that the writeLoop was spinning because another
routine was already in flushOutbound() and stack trace would
show that it was stuck in system calls. It seems that even though
the writeLoop does release the lock but grab it right away was
not allowing the syscall to complete.
So decided to put back the unlock/gosched/lock back in flushOutbound()
when flag is already set, but then protect the closeConnection()
with its own flag (similar to clearConnection) to not re-introduce
issue fixed in #1092.
Had to fix the benchmark test RoutedInterestGraph because after a
route is accepted, the initial PING will be sent after 1sec which
was breaking this test.
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
Remove sub from rsubs sublist when user UNSUBs.
Fix bench test that was not actually creating a SUB per request
in the Benchmark_Gateways_Requests_CreateOneSubForEach test.
Also UNSUBs older SUBs after a certain threshold to simulate
actual req/reply.
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
This addresses the following race:
- client connection creates a subscription on a reply subject
- client connection sends a request
- server sends the subscription to inbound gateway
- server sends the message to outbound gateway (those may be
to different servers)
- receiving server sends to sub interested in request subject
- app sends reply
- its server then check for interest on the reply's subject
In interestOnly mode, there is a possibility that this server
has not received the interest on the reply subject yet and would
then drop the reply.
This PR detects above scenario and will prefix the reply subject
to identify the origin cluster if it is detected that the last
subscription from the sending connection was created less than
a second ago.
Once the destination has this prefix, the destination cluster
will always send back that message to origin cluster even if
there is no registered interest.
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
Check sublist only when required.
Send the subs list in place instead of go routine (gateways have
different outbound/inbound connections so they don't suffer same
issue than routes)
Bump the default array size when collecting gateway connections
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
Allow deny clauses for subscriptions to still allow wildcard subscriptions but do not deliver the messages themselves.
Signed-off-by: Derek Collison <derek@nats.io>
